Start with AI

Two installs make an AI coding agent useful on VideoSDK.

The agent skill is the procedure. It works out what you are building, asks the handful of questions it needs, and mints auth tokens on a server route instead of leaving your API secret in browser code.

The MCP server is the lookup. It gives the agent the live documentation, so every method name it writes comes from the page you would have read yourself.

Quick start

1. Install the agent skill

npx skills add videosdk-live/skills

npx skills is agent-agnostic. It asks which agents to install into, so pick yours from the list (Claude Code, Cursor, Codex, and the others it supports). Restart the agent afterwards.

2. Connect the MCP server

The server is at https://docs.videosdk.live/mcp and speaks streamable HTTP. Pick your client:

Run this from your project, or add --scope user to make the server available everywhere:

claude mcp add --transport http videosdk https://docs.videosdk.live/mcp

Confirm it registered with claude mcp list.

Add the server to ~/.gemini/config/mcp_config.json for every project, or to .agents/mcp_config.json for this one:

{
  "mcpServers": {
    "videosdk": {
      "serverUrl": "https://docs.videosdk.live/mcp"
    }
  }
}

A remote server must use serverUrl. The legacy url and httpUrl fields are not supported.

Add the server to ~/.codex/config.toml:

[mcp_servers.videosdk]
url = "https://docs.videosdk.live/mcp"

Add the server to ~/.cursor/mcp.json for every project, or to .cursor/mcp.json for this one:

{
  "mcpServers": {
    "videosdk": {
      "url": "https://docs.videosdk.live/mcp"
    }
  }
}

Create .vscode/mcp.json in your workspace:

{
  "servers": {
    "videosdk": {
      "type": "http",
      "url": "https://docs.videosdk.live/mcp"
    }
  }
}

Copilot picks the server up once the file is saved.

Add the server to ~/.codeium/windsurf/mcp_config.json:

{
  "mcpServers": {
    "videosdk": {
      "serverUrl": "https://docs.videosdk.live/mcp"
    }
  }
}

Add the server under context_servers in your Zed settings.json:

{
  "context_servers": {
    "videosdk": {
      "source": "custom",
      "url": "https://docs.videosdk.live/mcp"
    }
  }
}

Open the MCP Servers panel, choose to edit the configuration, and add the server to cline_mcp_settings.json:

{
  "mcpServers": {
    "videosdk": {
      "type": "streamableHttp",
      "url": "https://docs.videosdk.live/mcp"
    }
  }
}

Most clients need a restart before a newly added server initialises. The MCP Server page covers verification and troubleshooting.

3. Prompt

Name the platform and say what the app does. Whatever you leave out comes back as a question.

Use the VideoSDK skill to add video calling to this Next.js app, with a server route that mints the auth token.

Build a telehealth consultation in Flutter, doctor and patient, with a waiting room.

Build a live shopping stream in React where viewers can be brought up on stage.

Check the setup worked

Watch the first task for these four. All four together mean both installs are live.

It asks before it builds. One round of scoping questions, then a one-sentence plan you can correct in a word.

It cites a real page. Ask where a method came from. The answer is a docs.videosdk.live URL it fetched during the task.

It keeps the API secret server side. Tokens get minted on a server route, and the client holds only the token.

It looks up the version you are on. Say the SDK line in the prompt, then check which page it cites. The server resolves the version, then searches inside it.

Skill or MCP server

They do different jobs and work best together.

	Agent Skill	MCP Server
What it is	The procedure the agent follows	The lookup the agent calls
What it holds	How to scope the build, wire auth, and pick features	The current documentation, searchable by SDK and version
When it runs	At the start of a VideoSDK task	Every time the agent needs an API detail

The skill decides what to build. When it needs a real method name, it asks the MCP server for the page.

Every SDK also has an AI Tools section in its own sidebar, with the library name the MCP server indexes and prompts written for that platform.

If your tool has no MCP support

Point it at the plain-text index instead.

https://docs.videosdk.live/llms.txt lists every page with its description.
https://docs.videosdk.live/llms-full.txt is the whole documentation set in one file.
Any page answers as raw markdown when you add .md to its route, so /react/guide/video-and-audio-calling-api-sdk/quick-start also serves /react/guide/video-and-audio-calling-api-sdk/quick-start.md.

Every page carries a copy button in the top right. It copies the cleaned markdown, or hands it to ChatGPT or Perplexity.

Start with AI

Quick start​

1. Install the agent skill​

2. Connect the MCP server​

3. Prompt​

Check the setup worked​

Skill or MCP server​

If your tool has no MCP support​